Katie Bauer!

Having spent much of my career working in social media or optimizing ad placement for social media sites, I figured I’d finally read Marshall McLuhan’s Understanding Media. I'm only about halfway through, but both the ideas and the writing itself are great

There is no universe where I’d be able to do justice to a book this dense by trying to summarize it in a Twitter thread. But I have been thinking about how it applies to data products, defined here as “things a data scientist might produce during the course of their work”

An idea introduced early in the book is that each medium of communication is a container for a previous form of communication. Thus, telegraphs contain print, print contains the written word, the written word contains speech, and speech contains thought itself.

You can apply this line of thinking to data products as well. Recommender systems predict the probability of a user clicking through an item in a feed. That prediction is a container for a summary stat, probably the mean of all possible outcomes given the input data.

That point estimate is in turn a container for a statistical distribution of those possible outcomes, which is a container for a statistical model that compresses the distributions for each input. And THOSE input distributions are containers for the data generating process

TL;DR it's statistical turtles all the way down

This may seem navel-gazey, but it’s getting at another of McLuhan’s points. We don't recognize new media when we first encounter it. We only see the old media format inside the new media container, so the new format alters the way we think without us even realizing it's happening